Suggesting features for AI Generative fill

[Encartauk]

- Have the option for the item generated to be cutout by default e.g. If I have ask for a flying pig, have it cutout so I can place it on any background/ move it around, etc.

- Have a 'flip' button for horizontal and verticle orientation so after the image is generated the user can quickly flip it (this wouldn't be need as much if the image was cutout like suggested in the first point.)

- Have a slider to adjust the blur/focus of the generated image so it blends in more acurately upon creation (again, maybe not such an issue if image was cutout already.)

- Have a feature where the user can bring in their own additional image and have it easily blend in with the existing image as the AI generation would have done e.g. If I want to place a hat on a gorilla, I can select my own hat image and the AI positions, blends, resizes, lighting, etc onto the gorilla as the AI may have done (I say 'may have done' as I'm sure I read Adobe claims these things get considered when it's AI places their generation.)

- Be able to just start typing a promt without selecting an area of the image. I have a picture of an animal behind a out-of-focus wire fence and wanted to see if AI would remove the fence and recreate the animal details where missing. I had to select the whole picture to be able to ask 'remove fence' in which it just replaced the whole picture with something irrelevant. So I then tried selecting just the fence (which was time consuming) and then asked again, only for the AI to basically modify the fence without removing it and messed up the animal behind. If I'm selecting just the fence I'm not sure the AI sees it as a fence or just grey blurry pixels, and then also whether the animal is considered too, despite being out of the selection. Not having to select an area first, unless necessary, will clear up any confusion about whether I should be spending a potentially large amount of time selecting areas first.

- Don't be so strict with the prohibited words. I've already read a huge post with people complaining about this, but I'll add one of my examples here. I asked for a picture of a jet fighter plane as part of a tutorial provided by Adobe of a lizard, then afterwards, asked for another prompt for 'jet thrusters' to show the exhaust flames etc that comes out the back. It got blocked. I guess because of the word 'thrust'! The word word thrust isn't offensive or inappropriate and can be applied to many movements not involving carnal activities, and even so, if it's a 'physical movement', even a lewd one, how can that be conveyed in a still image? And also the AI should be smart enough to be aware that the context of 'jet' was a huge factor.

This is all I have for now after a few hours of messing around with it. Occasionally it's impressive, mostly it's frustrating and I wonder if it's actually saving me (an experienced PS user - may be great for those not too proficient) any time over doing things the previous way.

[Frank22550268u9bs]

I'd like the generative fill to be able to create elements with a transparent background.  For example, you could request a campfire, chose the option that you like best and then move or scale it within your image.  Or add some smoke and then play with the placement and opacity.

"with transparent background" would be a great term for the system.  Here are some examples.

Add the trunk of a pine tree with a few broken branches with transparent background

Add small puffy clouds with transparent background

Add the front end of a red sports car coming in to the right side of the image with transparent background

[Encartauk]

The reason I mention a 'flip' button is because I asked for a dragonfly and a jet fighter plane and on both occasions I asked in the prompt to be 'facing left' / 'looking left' / 'flying left' , etc. and no matter what I tried, only about 10% of the time was the result facing left. Maybe I'm missing something but this is quite crucial to composing images that the parts are facing where you want them to be. And because the items, e.g. dragonfly, aren't cutout by default, if you flip them in the edit>transform>flip manner the surrounding opaque pixels get flipped too and the result doesn't blend in, so you have to spend extra time cutting them out (because Adobe's auto tools for that don't always have the best results either.)

[Conrad C]

This was originally posted in the community for the release version of Photoshop; suggestions for the beta version should go into the Ideas section of the Photoshop (Beta) community. Maybe a moderator can move this thread there because there are a lot of good ideas in it.

[davescm]

Hi , I've moved your post from the Photoshop forum to the Photoshop Beta forum (under 'Ideas') where you are more likely to get a response to your suggestions.

Dave

[Encartauk]

I just clicked the resulting link within Photoshop Beta's help section, after searching for 'feedback' as I had seen it mentioned somewhere within the software at some point and it took me here. All the text along the way suggested it was in the Beta section. Maybe that needs sorting too?

[Encartauk]

Yep, that's what my 1st point was on original post. I just used the term cut-out instead of transparent. I just think a tickbox for 'cutout' / 'isolated' etc would be enough.

However, I may have slightly misinterpreted your comment, I think? I'm thinking you mean, in your example with smoke or clouds, that the item itself would have transparency. That would be a useful feature too.

[cqqkiecqqkie]

and... hair!!

[gaylee0D45]

Id like to see the generative AI change the background without altering my subject at all even when the subject is selected it often deletes or changes part of it which makes it pretty useless for me.